Hwang's ResearchHwang's Research
Research

Seven areas of active investigation.

From distributed learning to quantum simulation — the questions we keep coming back to, and how we pursue them.

Federated Learning

distributedprivacyoptimizationcontinual

We train models across many clients without raw data leaving the device. Our work focuses on adaptive aggregation under heterogeneous data, continual and incremental client participation, and communication-efficient protocols that scale to thousands of edge nodes. We treat federated learning as a systems problem as much as an optimization one.

Autonomous Driving

perceptionplanningSSMMamba

End-to-end autonomous driving stacks built on modern sequence models. We're exploring Mamba-based state-space architectures as a more efficient alternative to attention for long-horizon perception and planning — preserving accuracy at a fraction of the compute cost, which is what real on-vehicle deployment demands.

Quantum Computing

algorithmssimulationvariationalhybrid

Quantum algorithm prototyping and circuit simulation for the era of hybrid classical-quantum systems. Our PRISM simulator focuses on making it easy to express and test variational and gate-based algorithms before deploying to real hardware, with a workflow geared toward researchers rather than vendors.

Related Projects

AI Systems

systemsinferencetrainingscale

The infrastructure layer of modern AI — scalable training, inference engines, and serving platforms. We work on the systems plumbing that lets research move at the speed of ideas rather than the speed of GPUs.

AI Agents

LLMtoolsreasoningevaluation

Multi-agent systems that combine foundation models with tools, memory, and structured reasoning. We're interested in what changes when agents can call into each other and operate over longer time horizons than a single chat turn — including the evaluation problem that comes with it.

Security

adversarialprivacyverification

Robust and secure machine learning — adversarial resilience, secure aggregation in federated settings, and verifiable privacy guarantees for systems that handle sensitive data. Security as a first-class research concern, not an afterthought.

Edge AI

compressionquantizationefficiency

On-device intelligence under tight compute, memory, and energy constraints. Model compression, quantization, and architecture search for hardware that runs on milliwatts — the kind of work that decides whether an idea actually deploys.